agora inbox for [email protected]help / color / mirror / Atom feed
RE: In-placre persistance change of a relation 7725+ messages / 3 participants [nested] [flat]
* RE: In-placre persistance change of a relation @ 2020-12-04 07:49 [email protected] <[email protected]> 2020-12-24 08:02 ` Re: In-placre persistance change of a relation Kyotaro Horiguchi <[email protected]> 0 siblings, 1 reply; 7725+ messages in thread From: [email protected] @ 2020-12-04 07:49 UTC (permalink / raw) To: 'Kyotaro Horiguchi' <[email protected]>; +Cc: [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]> From: Kyotaro Horiguchi <[email protected]> > > No, not really. The issue is more around what happens if we crash > > part way through. At crash recovery time, the system catalogs are not > > available, because the database isn't consistent yet and, anyway, the > > startup process can't be bound to a database, let alone every database > > that might contain unlogged tables. So the sentinel that's used to > > decide whether to flush the contents of a table or index is the > > presence or absence of an _init fork, which the startup process > > obviously can see just fine. The _init fork also tells us what to > > stick in the relation when we reset it; for a table, we can just reset > > to an empty file, but that's not legal for indexes, so the _init fork > > contains a pre-initialized empty index that we can just copy over. > > > > Now, to make an unlogged table logged, you've got to at some stage > > remove those _init forks. But this is not a transactional operation. > > If you remove the _init forks and then the transaction rolls back, > > you've left the system an inconsistent state. If you postpone the > > removal until commit time, then you have a problem if it fails, > > It's true. That are the cause of headache. ... > The current implement is simple. It's enough to just discard old or > new relfilenode according to the current transaction ends with commit > or abort. Tweaking of relfilenode under use leads-in some skews in > some places. I used pendingDelete mechanism a bit complexified way > and a violated an abstraction (I think, calling AM-routines from > storage.c is not good.) and even introduce a new fork kind only to > mark a init fork as "not committed yet". There might be better way, > but I haven't find it. I have no alternative idea yet, too. I agree that we want to avoid them, especially introducing inittmp fork... Anyway, below are the rest of my review comments for 0001. I want to review 0002 when we have decided to go with 0001. (2) XLOG_SMGR_UNLINK seems to necessitate modification of the following comments: [src/include/catalog/storage_xlog.h] /* * Declarations for smgr-related XLOG records * * Note: we log file creation and truncation here, but logging of deletion * actions is handled by xact.c, because it is part of transaction commit. */ [src/backend/access/transam/README] 3. Deleting a table, which requires an unlink() that could fail. Our approach here is to WAL-log the operation first, but to treat failure of the actual unlink() call as a warning rather than error condition. Again, this can leave an orphan file behind, but that's cheap compared to the alternatives. Since we can't actually do the unlink() until after we've committed the DROP TABLE transaction, throwing an error would be out of the question anyway. (It may be worth noting that the WAL entry about the file deletion is actually part of the commit record for the dropping transaction.) (3) +/* This is bit-map, not ordianal numbers */ There seems to be no comments using "bit-map". "Flags for ..." can be seen here and there. (4) Some wrong spellings: + /* we flush this buffer when swithing to PERMANENT */ swithing -> switching + * alredy flushed out by RelationCreate(Drop)InitFork called just alredy -> already + * relation content to be WAL-logged to recovery the table. recovery -> recover + * The inittmp fork works as the sentinel to identify that situaton. situaton -> situation (5) + table_close(classRel, NoLock); + + + + } These empty lines can be deleted. (6) +/* + * Perform XLogInsert of an XLOG_SMGR_UNLINK record to WAL. + */ +void +log_smgrbufpersistence(const RelFileNode *rnode, bool persistence) ... + * Make an XLOG entry reporting the file unlink. Not unlink but buffer persistence? (7) + /* + * index-init fork needs further initialization. ambuildempty shoud do + * WAL-log and file sync by itself but otherwise we do that by myself. + */ + if (rel->rd_rel->relkind == RELKIND_INDEX) + rel->rd_indam->ambuildempty(rel); + else + { + log_smgrcreate(&rnode, INIT_FORKNUM); + smgrimmedsync(srel, INIT_FORKNUM); + } + + /* + * We have created the init fork. If server crashes before the current + * transaction ends the init fork left alone corrupts data while recovery. + * The inittmp fork works as the sentinel to identify that situaton. + */ + smgrcreate(srel, INITTMP_FORKNUM, false); + log_smgrcreate(&rnode, INITTMP_FORKNUM); + smgrimmedsync(srel, INITTMP_FORKNUM); If the server crashes between these two processings, only the init fork exists. Is it correct to create the inittmp fork first? (8) + if (inxact_created) + { + SMgrRelation srel = smgropen(rnode, InvalidBackendId); + smgrclose(srel); + log_smgrunlink(&rnode, INIT_FORKNUM); + smgrunlink(srel, INIT_FORKNUM, false); + log_smgrunlink(&rnode, INITTMP_FORKNUM); + smgrunlink(srel, INITTMP_FORKNUM, false); + return; + } smgrclose() should be called just before return. Isn't it necessary here to revert buffer persistence state change? (9) +void +smgrunlink(SMgrRelation reln, ForkNumber forknum, bool isRedo) +{ + smgrsw[reln->smgr_which].smgr_unlink(reln->smgr_rnode, forknum, isRedo); +} Maybe it's better to restore smgrdounlinkfork() that was removed in the older release. That function includes dropping shared buffers, which can clean up the shared buffers that may be cached by this transaction. (10) [RelationDropInitFork] + /* revert buffer-persistence changes at abort */ + pending = (PendingRelDelete *) + MemoryContextAlloc(TopMemoryContext, sizeof(PendingRelDelete)); + pending->relnode = rnode; + pending->op = PDOP_SET_PERSISTENCE; + pending->bufpersistence = false; + pending->backend = InvalidBackendId; + pending->atCommit = true; + pending->nestLevel = GetCurrentTransactionNestLevel(); + pending->next = pendingDeletes; + pendingDeletes = pending; +} bufpersistence should be true. (11) + BlockNumber block = 0; ... + DropRelFileNodeBuffers(rbnode, &pending->unlink_forknum, 1, + &block); "block" is unnecessary and 0 can be passed directly. (12) - && pending->backend == InvalidBackendId) + && pending->backend == InvalidBackendId && + pending->op == PDOP_DELETE) nrels++; It's better to put && at the beginning of the line to follow the existing code here. (13) + table_close(rel, lockmode); lockmode should be NoLock to retain the lock until transaction completion. (14) + ctl.keysize = sizeof(unlogged_relation_entry); + ctl.entrysize = sizeof(unlogged_relation_entry); + hash = hash_create("unlogged hash", 32, &ctl, HASH_ELEM); ... + memset(key.oid, 0, sizeof(key.oid)); + memcpy(key.oid, de->d_name, oidchars); + ent = hash_search(hash, &key, HASH_FIND, NULL); keysize should be the oid member of the struct. Regards Takayuki Tsunakawa ^ permalink raw reply [nested|flat] 7725+ messages in thread
* Re: In-placre persistance change of a relation 2020-12-04 07:49 RE: In-placre persistance change of a relation [email protected] <[email protected]> @ 2020-12-24 08:02 ` Kyotaro Horiguchi <[email protected]> 2020-12-25 00:12 ` Re: In-placre persistance change of a relation Kyotaro Horiguchi <[email protected]> 0 siblings, 1 reply; 7725+ messages in thread From: Kyotaro Horiguchi @ 2020-12-24 08:02 UTC (permalink / raw) To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected] Thanks for the comment! Sorry for the late reply. At Fri, 4 Dec 2020 07:49:22 +0000, "[email protected]" <[email protected]> wrote in > From: Kyotaro Horiguchi <[email protected]> > > > No, not really. The issue is more around what happens if we crash > > > part way through. At crash recovery time, the system catalogs are not > > > available, because the database isn't consistent yet and, anyway, the > > > startup process can't be bound to a database, let alone every database > > > that might contain unlogged tables. So the sentinel that's used to > > > decide whether to flush the contents of a table or index is the > > > presence or absence of an _init fork, which the startup process > > > obviously can see just fine. The _init fork also tells us what to > > > stick in the relation when we reset it; for a table, we can just reset > > > to an empty file, but that's not legal for indexes, so the _init fork > > > contains a pre-initialized empty index that we can just copy over. > > > > > > Now, to make an unlogged table logged, you've got to at some stage > > > remove those _init forks. But this is not a transactional operation. > > > If you remove the _init forks and then the transaction rolls back, > > > you've left the system an inconsistent state. If you postpone the > > > removal until commit time, then you have a problem if it fails, > > > > It's true. That are the cause of headache. > ... > > The current implement is simple. It's enough to just discard old or > > new relfilenode according to the current transaction ends with commit > > or abort. Tweaking of relfilenode under use leads-in some skews in > > some places. I used pendingDelete mechanism a bit complexified way > > and a violated an abstraction (I think, calling AM-routines from > > storage.c is not good.) and even introduce a new fork kind only to > > mark a init fork as "not committed yet". There might be better way, > > but I haven't find it. > > I have no alternative idea yet, too. I agree that we want to avoid them, especially introducing inittmp fork... Anyway, below are the rest of my review comments for 0001. I want to review 0002 when we have decided to go with 0001. > > > (2) > XLOG_SMGR_UNLINK seems to necessitate modification of the following comments: > > [src/include/catalog/storage_xlog.h] > /* > * Declarations for smgr-related XLOG records > * > * Note: we log file creation and truncation here, but logging of deletion > * actions is handled by xact.c, because it is part of transaction commit. > */ Sure. Rewrote it. > [src/backend/access/transam/README] > 3. Deleting a table, which requires an unlink() that could fail. > > Our approach here is to WAL-log the operation first, but to treat failure > of the actual unlink() call as a warning rather than error condition. > Again, this can leave an orphan file behind, but that's cheap compared to > the alternatives. Since we can't actually do the unlink() until after > we've committed the DROP TABLE transaction, throwing an error would be out > of the question anyway. (It may be worth noting that the WAL entry about > the file deletion is actually part of the commit record for the dropping > transaction.) Mmm. I didn't touched theDROP TABLE (RelationDropStorage) path, but I added a brief description about INITTMP fork to the file. ==== The INITTMP fork file -------------------------------- An INITTMP fork is created when new relation file is created to mark the relfilenode needs to be cleaned up at recovery time. The file is removed at transaction end but is left when the process crashes before the transaction ends. In contrast to 4 above, failure to remove an INITTMP file will lead to data loss, in which case the server will shut down. ==== > (3) > +/* This is bit-map, not ordianal numbers */ > > There seems to be no comments using "bit-map". "Flags for ..." can be seen here and there. I revmoed the comment and use (1 << n) notation to show the fact instead. > (4) > Some wrong spellings: > > swithing -> switching > alredy -> already > recovery -> recover > situaton -> situation Oops! Fixed them. > (5) > + table_close(classRel, NoLock); > + > + > + > + > } > > These empty lines can be deleted. s/can/should/ :p. Fixed. > > (6) > +/* > + * Perform XLogInsert of an XLOG_SMGR_UNLINK record to WAL. > + */ > +void > +log_smgrbufpersistence(const RelFileNode *rnode, bool persistence) > ... > + * Make an XLOG entry reporting the file unlink. > > Not unlink but buffer persistence? Silly copy-pasto. Fixed. > (7) > + /* > + * index-init fork needs further initialization. ambuildempty shoud do > + * WAL-log and file sync by itself but otherwise we do that by myself. > + */ > + if (rel->rd_rel->relkind == RELKIND_INDEX) > + rel->rd_indam->ambuildempty(rel); > + else > + { > + log_smgrcreate(&rnode, INIT_FORKNUM); > + smgrimmedsync(srel, INIT_FORKNUM); > + } > + > + /* > + * We have created the init fork. If server crashes before the current > + * transaction ends the init fork left alone corrupts data while recovery. > + * The inittmp fork works as the sentinel to identify that situaton. > + */ > + smgrcreate(srel, INITTMP_FORKNUM, false); > + log_smgrcreate(&rnode, INITTMP_FORKNUM); > + smgrimmedsync(srel, INITTMP_FORKNUM); > > If the server crashes between these two processings, only the init fork exists. Is it correct to create the inittmp fork first? Right. I change it that way, and did the same with the new code added to RelationCreateStorage. > (8) > + if (inxact_created) > + { > + SMgrRelation srel = smgropen(rnode, InvalidBackendId); > + smgrclose(srel); > + log_smgrunlink(&rnode, INIT_FORKNUM); > + smgrunlink(srel, INIT_FORKNUM, false); > + log_smgrunlink(&rnode, INITTMP_FORKNUM); > + smgrunlink(srel, INITTMP_FORKNUM, false); > + return; > + } > > smgrclose() should be called just before return. > Isn't it necessary here to revert buffer persistence state change? Mmm. it's a thinko. I was confused with the case of close/unlink. Fixed all instacnes of the same. > (9) > +void > +smgrunlink(SMgrRelation reln, ForkNumber forknum, bool isRedo) > +{ > + smgrsw[reln->smgr_which].smgr_unlink(reln->smgr_rnode, forknum, isRedo); > +} > > Maybe it's better to restore smgrdounlinkfork() that was removed in the older release. That function includes dropping shared buffers, which can clean up the shared buffers that may be cached by this transaction. INITFORK/INITTMP forks cannot be loaded to shared buffer so it's no use to drop buffers. I added a comment like that. | /* | * INIT/INITTMP forks never be loaded to shared buffer so no point in | * dropping buffers for these files. | */ | log_smgrunlink(&rnode, INIT_FORKNUM); I removed DropRelFileNodeBuffers from PDOP_UNLINK_FORK branch in smgrDoPendingDeletes and added an assertion and a comment instead. | /* other forks needs to drop buffers */ | Assert(pending->unlink_forknum == INIT_FORKNUM || | pending->unlink_forknum == INITTMP_FORKNUM); | | log_smgrunlink(&pending->relnode, pending->unlink_forknum); | smgrunlink(srel, pending->unlink_forknum, false); > (10) > [RelationDropInitFork] > + /* revert buffer-persistence changes at abort */ > + pending = (PendingRelDelete *) > + MemoryContextAlloc(TopMemoryContext, sizeof(PendingRelDelete)); > + pending->relnode = rnode; > + pending->op = PDOP_SET_PERSISTENCE; > + pending->bufpersistence = false; > + pending->backend = InvalidBackendId; > + pending->atCommit = true; > + pending->nestLevel = GetCurrentTransactionNestLevel(); > + pending->next = pendingDeletes; > + pendingDeletes = pending; > +} > > bufpersistence should be true. RelationDropInitFork() chnages the relation persisitence to "persistent" so it shoud be reverted to "non-persistent (= false)" at abort. (I agree that the function name is somewhat confusing...) > (11) > + BlockNumber block = 0; > ... > + DropRelFileNodeBuffers(rbnode, &pending->unlink_forknum, 1, > + &block); > > "block" is unnecessary and 0 can be passed directly. I removed the entire function call. But, I don't think you're right here. | DropRelFileNodeBuffers(RelFileNodeBackend rnode, ForkNumber *forkNum, | int nforks, BlockNumber *firstDelBlock) Doesn't just passing 0 lead to SEGV? > (12) > - && pending->backend == InvalidBackendId) > + && pending->backend == InvalidBackendId && > + pending->op == PDOP_DELETE) > nrels++; > > It's better to put && at the beginning of the line to follow the existing code here. It's terrible.. Fixed. > (13) > + table_close(rel, lockmode); > > lockmode should be NoLock to retain the lock until transaction completion. I tried to recall the reason for that, but didn't come up with anything. Fixed. > (14) > + ctl.keysize = sizeof(unlogged_relation_entry); > + ctl.entrysize = sizeof(unlogged_relation_entry); > + hash = hash_create("unlogged hash", 32, &ctl, HASH_ELEM); > ... > + memset(key.oid, 0, sizeof(key.oid)); > + memcpy(key.oid, de->d_name, oidchars); > + ent = hash_search(hash, &key, HASH_FIND, NULL); > > keysize should be the oid member of the struct. It's not a problem since the first member is the oid and perhaps it seems that I thougth to do someting more on that. Now that I don't recall what is it and in the first place the key should be just Oid in the context above. Fixed. The patch is attached to the next message. regards. -- Kyotaro Horiguchi NTT Open Source Software Center ^ permalink raw reply [nested|flat] 7725+ messages in thread
* Re: In-placre persistance change of a relation 2020-12-04 07:49 RE: In-placre persistance change of a relation [email protected] <[email protected]> 2020-12-24 08:02 ` Re: In-placre persistance change of a relation Kyotaro Horiguchi <[email protected]> @ 2020-12-25 00:12 ` Kyotaro Horiguchi <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Kyotaro Horiguchi @ 2020-12-25 00:12 UTC (permalink / raw) To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected] Hello. At Thu, 24 Dec 2020 17:02:20 +0900 (JST), Kyotaro Horiguchi <[email protected]> wrote in > The patch is attached to the next message. The reason for separating this message is that I modified this so that it could solve another issue. There's a complain about orphan files after crash. [1] 1: https://www.postgresql.org/message-id/[email protected] That is, the case where a relation file is left alone after a server crash that happened before the end of the transaction that has created a relation. As I read this, I noticed this feature can solve the issue with a small change. This version gets changes in RelationCreateStorage and smgrDoPendingDeletes. Previously inittmp fork is created only along with an init fork. This version creates one always when a relation storage file is created. As the result ResetUnloggedRelationsInDbspaceDir removes all forks if the inttmp fork of a logged relations is found. Now that pendingDeletes can contain multiple entries for the same relation, it has been modified not to close the same smgr multiple times. - It might be better to split 0001 into two peaces. - The function name ResetUnloggedRelationsInDbspaceDir is no longer represents the function correctly. regards. -- Kyotaro Horiguchi NTT Open Source Software Center ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 065316ee866..77c206ff944 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1084,6 +1084,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --nzintiedl6o4kcyp-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 140 insertions(+), 23 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 018e9f7b013..423cea26b0b 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1075,6 +1075,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..b02ec4e8c3e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..c785a3db59f 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1370,9 +1373,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index 1dad125706e..c6693b93a65 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -69,10 +69,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --gwom7bl7ogtszo4k-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/cluster.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 73 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 6 +- src/include/access/xlog_internal.h | 2 +- src/include/storage/proc.h | 3 +- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 10 files changed, 138 insertions(+), 22 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..79c052d3607 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXid %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 4fd10a5e7b3..73407c383da 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1086,6 +1086,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index 0f913897acc..f1a03e6a754 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,13 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2687,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2702,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2711,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2732,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2754,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2773,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2826,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2848,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index f3ad90c7c7a..5c4121fcbc8 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1358,10 +1359,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1372,7 +1375,8 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) /* array of TransactionIds */ if (xlrec.xcnt > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 629ac3a7d3e..1730b07810f 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -31,7 +31,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11D /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h index bf3094f0f7d..120370f459d 100644 --- a/src/include/storage/proc.h +++ b/src/include/storage/proc.h @@ -67,10 +67,11 @@ struct XidCache #define PROC_AFFECTS_ALL_HORIZONS 0x20 /* this proc's xmin must be * included in vacuum horizons * in all databases */ +#define PROC_IN_REPACK 0x40 /* currently REPACK (CONCURRENTLY) */ /* flags reset at EOXact */ #define PROC_VACUUM_STATE_MASK \ - (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND) + (PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | PROC_IN_REPACK) /* * Xmin-related flags. Make sure any flags that affect how the process' Xmin diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index c63a4f2cc6a..0bd28e270f1 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -124,10 +124,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --=-=-=-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 7291b4180ab..0103bc10bd1 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --cldir6umqisr673d-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 781a471fc1d..29ba49744eb 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..f2008c6877c 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --qfkt2ktdpcfeypib-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 3acb2e1d2ba..74faa4f5582 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --2fxmamo6mu2qbxgv-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 38 ++++++++--- src/backend/storage/ipc/procarray.c | 74 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 3 + src/include/storage/standbydefs.h | 2 + 8 files changed, 137 insertions(+), 21 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..b96b73d51d3 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index f5c9a34374d..d050b0b1444 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8477,6 +8478,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 21ab81c6f7f..42165c5552e 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -988,6 +988,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is + * o.k. for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 9cf499ce7c6..fbdd4600a2b 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This * means all transactions starting afterwards have enough information to * be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is + * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXidLogical is * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all * transactions that are currently in progress have a catalog snapshot, * and all their changes have been collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1459,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..6dd4fa7dcc0 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,24 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2672,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2688,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2703,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2712,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2733,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2755,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2774,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2827,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2849,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..9e6e7b18786 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,13 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK + * (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --mdsv5bdgfjjj6d77-- ^ permalink raw reply [nested|flat] 7725+ messages in thread
* [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). @ 2026-03-17 19:22 Antonin Houska <[email protected]> 0 siblings, 0 replies; 7725+ messages in thread From: Antonin Houska @ 2026-03-17 19:22 UTC (permalink / raw) During logical decoding, we need to know if particular transaction performed catalog changes because catalog information is needed to construct heap tuples. To be sure that we have enough information of each transaction, the logical decoding cannot start before all the already running transactions have completed. The problem with REPACK (CONCURRENTLY) is that it has XID assigned and writes WAL records marked with it. Thus if another backend runs REPACK (CONCURRENTLY) and tries to setup the logical decoding, it has to wait for the completion of all the other transactions involved in REPACK (CONCURRENTLY). However, REPACK (CONCURRENTLY) does not perform any catalog changes relevant to logical decoding, so the other backends executing this command can ignore it. This patch implements it by adding information about transactions executing the command to the xl_running_xacts WAL record, and by teaching the snapshot builder to use the information. --- src/backend/access/rmgrdesc/standbydesc.c | 15 ++++- src/backend/access/transam/xlog.c | 2 + src/backend/commands/repack.c | 16 +++++ src/backend/replication/logical/snapbuild.c | 51 +++++++++----- src/backend/storage/ipc/procarray.c | 75 +++++++++++++++++++-- src/backend/storage/ipc/standby.c | 8 ++- src/include/storage/standby.h | 2 + src/include/storage/standbydefs.h | 2 + 8 files changed, 144 insertions(+), 27 deletions(-) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..58ebecf4927 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -21,10 +21,11 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) { int i; - appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u", + appendStringInfo(buf, "nextXid %u latestCompletedXid %u oldestRunningXid %u oldestRunningXidLogical %u", xlrec->nextXid, xlrec->latestCompletedXid, - xlrec->oldestRunningXid); + xlrec->oldestRunningXid, + xlrec->oldestRunningXidLogical); if (xlrec->xcnt > 0) { appendStringInfo(buf, "; %d xacts:", xlrec->xcnt); @@ -41,6 +42,16 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + if (xlrec->xcnt_repack > 0) + { + TransactionId *xids_repack; + + appendStringInfo(buf, "; %d xacts_repack:", xlrec->xcnt_repack); + xids_repack = xlrec->xids + xlrec->xcnt + xlrec->subxcnt; + for (i = 0; i < xlrec->xcnt_repack; i++) + appendStringInfo(buf, " %u", xids_repack[i]); + } } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 2c1c6f88b74..f1cbdee7618 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -5912,6 +5912,7 @@ StartupXLOG(void) * subxids are listed with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); @@ -8483,6 +8484,7 @@ xlog_redo(XLogReaderState *record) * with their parent prepared transactions. */ running.xcnt = nxids; + running.xcnt_repack = 0; running.subxcnt = 0; running.subxid_status = SUBXIDS_IN_SUBTRANS; running.nextXid = XidFromFullTransactionId(checkPoint.nextXid); diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index 274ad900c65..5188244dfd6 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -993,6 +993,22 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, if (concurrent) { + /* + * Do not let other backends wait for our completion during their + * setup of logical replication. Unlike logical replication publisher, + * we will have XID assigned, so the other backends - whether + * walsenders involved in logical replication or regular backends + * executing also REPACK (CONCURRENTLY) - would have to wait for our + * completion before they can build their initial snapshot. It is o.k. + * for any decoding backend to ignore us because we do not change + * tuple descriptor of any table, and the data changes we write should + * not be decoded by other backends. + */ + LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE); + MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK; + ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags; + LWLockRelease(ProcArrayLock); + /* * The worker needs to be member of the locking group we're the leader * of. We ought to become the leader before the worker starts. The diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index 0f6b5df322c..23511a0905a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -1172,7 +1172,7 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact * xmin, which looks odd but is correct and actually more efficient, since * we hit fast paths in heapam_visibility.c. */ - builder->xmin = running->oldestRunningXid; + builder->xmin = running->oldestRunningXidLogical; /* Remove transactions we don't need to keep track off anymore */ SnapBuildPurgeOlderTxn(builder); @@ -1188,9 +1188,9 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ xmin = ReorderBufferGetOldestXmin(builder->reorder); if (xmin == InvalidTransactionId) - xmin = running->oldestRunningXid; + xmin = running->oldestRunningXidLogical; elog(DEBUG3, "xmin: %u, xmax: %u, oldest running: %u, oldest xmin: %u", - builder->xmin, builder->xmax, running->oldestRunningXid, xmin); + builder->xmin, builder->xmax, running->oldestRunningXidLogical, xmin); LogicalIncreaseXminForSlot(lsn, xmin); /* @@ -1275,14 +1275,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * have all necessary catalog rows anymore. */ if (TransactionIdIsNormal(builder->initial_xmin_horizon) && - NormalTransactionIdPrecedes(running->oldestRunningXid, + NormalTransactionIdPrecedes(running->oldestRunningXidLogical, builder->initial_xmin_horizon)) { ereport(DEBUG1, errmsg_internal("skipping snapshot at %X/%08X while building logical decoding snapshot, xmin horizon too low", LSN_FORMAT_ARGS(lsn)), errdetail_internal("initial xmin horizon of %u vs the snapshot's %u", - builder->initial_xmin_horizon, running->oldestRunningXid)); + builder->initial_xmin_horizon, running->oldestRunningXidLogical)); SnapBuildWaitSnapshot(running, builder->initial_xmin_horizon); @@ -1299,7 +1299,7 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn * NB: We might have already started to incrementally assemble a snapshot, * so we need to be careful to deal with that. */ - if (running->oldestRunningXid == running->nextXid) + if (running->oldestRunningXidLogical == running->nextXid) { if (!XLogRecPtrIsValid(builder->start_decoding_at) || builder->start_decoding_at <= lsn) @@ -1378,14 +1378,14 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from BUILDING_SNAPSHOT to FULL_SNAPSHOT. * - * In BUILDING_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid - * is >= than nextXid from when we switched to BUILDING_SNAPSHOT. This - * means all transactions starting afterwards have enough information to - * be decoded. Switch to FULL_SNAPSHOT. + * In BUILDING_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * BUILDING_SNAPSHOT. This means all transactions starting afterwards + * have enough information to be decoded. Switch to FULL_SNAPSHOT. */ else if (builder->state == SNAPBUILD_BUILDING_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_FULL_SNAPSHOT; builder->next_phase_at = running->nextXid; @@ -1402,14 +1402,15 @@ SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *runn /* * c) transition from FULL_SNAPSHOT to CONSISTENT. * - * In FULL_SNAPSHOT state, and this xl_running_xacts' oldestRunningXid is - * >= than nextXid from when we switched to FULL_SNAPSHOT. This means all - * transactions that are currently in progress have a catalog snapshot, - * and all their changes have been collected. Switch to CONSISTENT. + * In FULL_SNAPSHOT state, and this xl_running_xacts' + * oldestRunningXidLogical is >= than nextXid from when we switched to + * FULL_SNAPSHOT. This means all transactions that are currently in + * progress have a catalog snapshot, and all their changes have been + * collected. Switch to CONSISTENT. */ else if (builder->state == SNAPBUILD_FULL_SNAPSHOT && TransactionIdPrecedesOrEquals(builder->next_phase_at, - running->oldestRunningXid)) + running->oldestRunningXidLogical)) { builder->state = SNAPBUILD_CONSISTENT; builder->next_phase_at = InvalidTransactionId; @@ -1459,6 +1460,24 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) if (TransactionIdFollows(xid, cutoff)) continue; + /* Do not wait for transactions running REPACK (CONCURRENTLY). */ + if (running->xcnt_repack > 0) + { + TransactionId *xids_repack; + int i; + + xids_repack = running->xids + running->xcnt + running->subxcnt; + + for (i = 0; i < running->xcnt_repack; i++) + { + if (xid == xids_repack[i]) + break; + } + /* Found? */ + if (i < running->xcnt_repack) + continue; + } + XactLockTableWait(xid, NULL, NULL, XLTW_None); } diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index cc207cb56e3..033ae68b57e 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2643,15 +2643,25 @@ GetRunningTransactionData(void) RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData; TransactionId latestCompletedXid; TransactionId oldestRunningXid; + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; TransactionId *xids; int index; - int count; + int count, + count_repack; int subcount; bool suboverflowed; + TransactionId *xids_repack = NULL; + bool logical_decoding_enabled = IsLogicalDecodingEnabled(); Assert(!RecoveryInProgress()); + /* + * TODO Consider a GUC to reserve certain amount of replication slots for + * REPACK (CONCURRENTLY) and using it here. + */ +#define MAX_REPACK_XIDS 16 + /* * Allocating space for maxProcs xids is usually overkill; numProcs would * be sufficient. But it seems better to do the malloc while not holding @@ -2663,11 +2673,14 @@ GetRunningTransactionData(void) */ if (CurrentRunningXacts->xids == NULL) { + /* FIXME probably fails if logical decoding is enable on-the-fly */ + int nrepack = logical_decoding_enabled ? MAX_REPACK_XIDS : 0; + /* * First call */ CurrentRunningXacts->xids = (TransactionId *) - malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId)); + malloc((TOTAL_MAX_CACHED_SUBXIDS + nrepack) * sizeof(TransactionId)); if (CurrentRunningXacts->xids == NULL) ereport(ERROR, (errcode(ERRCODE_OUT_OF_MEMORY), @@ -2676,7 +2689,10 @@ GetRunningTransactionData(void) xids = CurrentRunningXacts->xids; - count = subcount = 0; + if (logical_decoding_enabled) + xids_repack = palloc_array(TransactionId, MAX_REPACK_XIDS); + + count = subcount = count_repack = 0; suboverflowed = false; /* @@ -2688,7 +2704,7 @@ GetRunningTransactionData(void) latestCompletedXid = XidFromFullTransactionId(TransamVariables->latestCompletedXid); - oldestDatabaseRunningXid = oldestRunningXid = + oldestDatabaseRunningXid = oldestRunningXid = oldestRunningXidLogical = XidFromFullTransactionId(TransamVariables->nextXid); /* @@ -2697,6 +2713,8 @@ GetRunningTransactionData(void) for (index = 0; index < arrayP->numProcs; index++) { TransactionId xid; + int pgprocno; + PGPROC *proc; /* Fetch xid just once - see GetNewTransactionId */ xid = UINT32_ACCESS_ONCE(other_xids[index]); @@ -2716,6 +2734,21 @@ GetRunningTransactionData(void) if (TransactionIdPrecedes(xid, oldestRunningXid)) oldestRunningXid = xid; + if (logical_decoding_enabled && + TransactionIdPrecedes(xid, oldestRunningXidLogical)) + { + /* + * Backends running REPACK concurrently need to be excluded from + * oldestRunningXidLogical, otherwise the snapshot builder cannot + * proceed in building the initial snapshot. + */ + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if ((proc->statusFlags & PROC_IN_CONCURRENT_REPACK) == 0) + oldestRunningXidLogical = xid; + } + /* * Also, update the oldest running xid within the current database. As * fetching pgprocno and PGPROC could cause cache misses, we do cheap @@ -2723,8 +2756,8 @@ GetRunningTransactionData(void) */ if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid)) { - int pgprocno = arrayP->pgprocnos[index]; - PGPROC *proc = &allProcs[pgprocno]; + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; if (proc->databaseId == MyDatabaseId) oldestDatabaseRunningXid = xid; @@ -2742,6 +2775,19 @@ GetRunningTransactionData(void) */ xids[count++] = xid; + + /* + * Collect XIDSs of transactions running REPACK (CONCURRENTLY). + */ + if (logical_decoding_enabled && + count_repack < MAX_REPACK_XIDS) + { + pgprocno = arrayP->pgprocnos[index]; + proc = &allProcs[pgprocno]; + + if (proc->statusFlags & PROC_IN_CONCURRENT_REPACK) + xids_repack[count_repack++] = xid; + } } /* @@ -2782,6 +2828,19 @@ GetRunningTransactionData(void) } } + /* + * Append the XIDs running REPACK (CONCURRENTLY), if any. + * + * XXX Should we sort the array and use bsearch() when using it? + */ + if (count_repack > 0) + { + for (int i = 0; i < count_repack; i++) + xids[count++] = xids_repack[i]; + } + if (xids_repack) + pfree(xids_repack); + /* * It's important *not* to include the limits set by slots here because * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those @@ -2791,11 +2850,13 @@ GetRunningTransactionData(void) * increases if slots do. */ - CurrentRunningXacts->xcnt = count - subcount; + CurrentRunningXacts->xcnt = count - subcount - count_repack; CurrentRunningXacts->subxcnt = subcount; + CurrentRunningXacts->xcnt_repack = count_repack; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid); CurrentRunningXacts->oldestRunningXid = oldestRunningXid; + CurrentRunningXacts->oldestRunningXidLogical = oldestRunningXidLogical; CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid; CurrentRunningXacts->latestCompletedXid = latestCompletedXid; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..fd3e4fe31f3 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1189,6 +1189,7 @@ standby_redo(XLogReaderState *record) RunningTransactionsData running; running.xcnt = xlrec->xcnt; + running.xcnt_repack = xlrec->xcnt_repack; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; running.nextXid = xlrec->nextXid; @@ -1359,10 +1360,12 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRecPtr recptr; xlrec.xcnt = CurrRunningXacts->xcnt; + xlrec.xcnt_repack = CurrRunningXacts->xcnt_repack; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); xlrec.nextXid = CurrRunningXacts->nextXid; xlrec.oldestRunningXid = CurrRunningXacts->oldestRunningXid; + xlrec.oldestRunningXidLogical = CurrRunningXacts->oldestRunningXidLogical; xlrec.latestCompletedXid = CurrRunningXacts->latestCompletedXid; /* Header */ @@ -1371,9 +1374,10 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) XLogRegisterData(&xlrec, MinSizeOfXactRunningXacts); /* array of TransactionIds */ - if (xlrec.xcnt > 0) + if (xlrec.xcnt + xlrec.xcnt_repack > 0) XLogRegisterData(CurrRunningXacts->xids, - (xlrec.xcnt + xlrec.subxcnt) * sizeof(TransactionId)); + (xlrec.xcnt + xlrec.xcnt_repack + xlrec.subxcnt) * + sizeof(TransactionId)); recptr = XLogInsert(RM_STANDBY_ID, XLOG_RUNNING_XACTS); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..fe4dcdc3def 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -127,10 +127,12 @@ typedef enum typedef struct RunningTransactionsData { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY). */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId oldestDatabaseRunningXid; /* same as above, but within the * current database */ TransactionId latestCompletedXid; /* so we can set xmax */ diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..edad609fa9a 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -47,10 +47,12 @@ typedef struct xl_standby_locks typedef struct xl_running_xacts { int xcnt; /* # of xact ids in xids[] */ + int xcnt_repack; /* # of xacts running REPACK (CONCURRENTLY) */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ TransactionId nextXid; /* xid from TransamVariables->nextXid */ TransactionId oldestRunningXid; /* *not* oldestXmin */ + TransactionId oldestRunningXidLogical; TransactionId latestCompletedXid; /* so we can set xmax */ TransactionId xids[FLEXIBLE_ARRAY_MEMBER]; -- 2.47.3 --3n3vqr5y2jhknrqj Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v49-0007-WIP-add-max_repack_replication_slots.patch" ^ permalink raw reply [nested|flat] 7725+ messages in thread
end of thread, other threads:[~2026-03-17 19:22 UTC | newest] Thread overview: 7725+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2020-12-04 07:49 RE: In-placre persistance change of a relation [email protected] <[email protected]> 2020-12-24 08:02 ` Kyotaro Horiguchi <[email protected]> 2020-12-25 00:12 ` Kyotaro Horiguchi <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v43 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v44 10/10] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v45 8/8] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v48 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v47 9/9] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v46 7/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]> 2026-03-17 19:22 [PATCH v49 6/7] Teach snapshot builder to skip transactions running REPACK (CONCURRENTLY). Antonin Houska <[email protected]>
This inbox is served by agora; see mirroring instructions for how to clone and mirror all data and code used for this inbox